LDSC-SEG

LDSC-SEG consists of two separate steps. The first step of LDSC-SEG is a differential expression analysis on the gene expression data to identify a set of genes that are specifically expressed in certain tissues. These tissue specific genes are referred to either as specifically/differentially expressed genes (SEGs) or tissue-specific genes (TSGs). In the differential expression analysis, LDSC-SEG examines one gene at a time. For the given gene, LDSC-SEG contrasts the gene expression level of samples collected in a focal tissue (e.g., brain-cortex) with those of samples collected in all other tissues that are not in the same tissue category as the focal tissue (i.e., non-brain tissues). Because tissues within each tissue category tend to share similarly expressed genes, excluding the tissues in the same tissue category in the differential expression analysis step becomes the key to ensure robust detection of SEGs. Indeed, such differential expression analysis allows for the inclusion of as many genes as possible that are highly expressed in the focal tissues but not in tissues from other tissue categories. The SEG evidence for a gene is typically characterized by a t-statistic, with a higher value indicating that the gene is more specifically/differentially expressed in the focal tissue. With the differential expression analysis results, LDSC-SEG ranks all genes in a descending order based on their t-statistics. LDSC-SEG then defines SEGs as the top 10 percentage of all genes. The identification of SEGs allows LDSC-SEG to create a binary SNP annotation in a tissue specific fashion. In particular, for each tissue at a time, LDSC-SEG annotates the SNP to be one if the SNP resides within 100 kb of the transcription start site of any SEG and annotates it to be zero otherwise. With the tissue-specific binary annotation, LDSC-SEG then performs the second step of applying the sLDSC method described in the previous section to estimate the proportion of SNP heritability explained by each tissue-specific binary SNP annotation. The resulting test statistic from sLDSC is then served as a relevance evidence between the tissue and trait.

In real data applications, LDSC-SEG analyzed GWAS summary statistics for 48 diseases and traits and found significant tissue-/cell type-specific enrichments for 34 traits. Several of these findings recapitulate known biology. For example, immunological traits exhibit immune tissue-type enrichments; psychiatric traits exhibit strong brain-related tissue enrichments; and type II diabetes exhibits enrichments in the pancreas. LDSC-SEG also validated several recent genetic analyses results, including robust brain-specific enrichments for smoking status, years of education, body mass index, and age at menarche.